METHOD AND COMPUTER PROGRAMME FOR DISCOVERING POSSIBLE ERRORS IN A PRODUCTION PROCESS

Abstract

A method and a computer program for discovering possible errors in a production process for manufacturing metal products. The method involves removing at least one process parameter value from the cluster or adding at least one process parameter to the cluster. Second Z-score values, which are compared with the first Z-score values, are then determined for the thus-altered cluster. The changes in the Z-score values provide suggestions for troubleshooting and process optimization which are direct and can be implemented immediately.

Claims

1-10. (canceled)

11. A computer-supported method for discovering at least one possible error in a production process for creating a metal product in a metallurgical system, comprising: defining a plurality of process parameters representing the production process in a parameter space of the production process, detecting or specifying values for the process parameters, and grouping of a part of these process parameter values into at least one cluster in the parameter space; applying a performance enrichment analysis to the cluster and to at least one performance indicator assigned to the cluster in order to determine a first dependency between the cluster and the performance indicator, wherein the dependency is evaluated with a first Z-score value; wherein generating a modified cluster by removing from the cluster or adding to the cluster at least one process parameter value; repeated applying of the performance enrichment analysis to the modified cluster and the assigned performance indicator to determine a second Z-score value between the modified cluster and the assigned performance indicator; and determining a change between the first Z-score value and the second Z-score value as a measure of the influence of the process parameter value removed from the cluster or added to the cluster on the dependency between the performance indicator and the cluster and thus as an indicator of a possible error and the cause thereof.

12. The method according to claim 11, wherein defining a threshold value for the change between the first and the second Z-score value, wherein the influence of the removed or added parameter value on the dependency between the performance indicator and the cluster is considered to be significant when the change exceeds the threshold value.

13. A computer-supported method for discovering at least one possible error in a production process for creating a metal product in a metallurgical system, having the following steps: defining a plurality of process parameters representing the production process in a parameter space of the production process, detecting or specifying values for the process parameters, and grouping of a part of these process parameter values into at least one cluster in the parameter space; applying a performance enrichment analysis to the cluster and to at least one external process parameter value of the parameter space not assigned to the cluster in order to determine a first dependency between the cluster and the external process parameter value, wherein the dependency is evaluated with a first Z-score value; wherein generating a modified cluster by removing from the cluster or adding to the cluster at least one process parameter value; and repeated applying of the performance enrichment analysis to the modified cluster and the external process parameter value to determine a second Z-score value between the modified cluster and the external process parameter value; and determining the change between the first Z-score value and the second Z-score value as a measure of the influence of the process parameter value removed from the cluster or added to the cluster on the dependency between the cluster and the external process parameter value, and thus as an indicator of a possible error and the cause thereof.

14. The method according to claim 13, wherein defining a threshold value for the change between the first and the second Z-score value, wherein the influence of the removed or added parameter value from the parameter space on the dependency between the cluster and the external process parameter value is considered to be significant when the change exceeds the threshold value.

15. The method according to claim 11, wherein the dimension of the parameter space and/or of the cluster is reduced, for example by eliminating highly correlated process parameters or by transforming vectors of the process parameters to main components within the scope of a principal component analysis (PCA), before the applying of the performance enrichment analysis.

16. The method according to claim 11, wherein visualizing the result of the first performance enrichment analysis in the form of a first dependency network, in which the process parameters from the parameter space or the clusters formed from the process parameter values, on the one hand, and the performance indicators or the external process parameter values, on the other hand, respectively form the nodes of the dependency network, and in which the connections between the nodes, evaluated with the Z-score value, represent the influence of the removed or added process parameter value on the performance indicator or the external process parameter value.

17. The method according to claim 16, wherein additional illustrating of the results of the repeated second performance enrichment analysis in the form of a second dependency network; and combining the first and the second dependency network into one overall network with enrichment-based relationships between the clusters and the performance indicators or the external process parameter values.

18. The method according to claim 11, wherein the performance indicator of the production process involve the following attributes: quality devaluations of the created metal product or scrap quantities; on-time schedule variances; and output quantities.

19. The method according to claim 11, wherein the process parameters from the parameter space, within and outside of the clusters, may be, for example, the following parameters: as relates to the type of product: the metal quality or the product category; as relates to the dimensions of the product: width, thickness; length, format as relates to the weight: weight per piece, coil weight, coating weight; as relates to the production capacity: mass flow, quantity; as relates to the mode of operation of the metallurgical system or one of the components thereof: pass schedule, speed.

20. A computer program which can be directly loaded into the internal memory of a digital computer and comprises sections of software code, with which the steps according to claim 11 are executed when the computer program is running on the computer.

21. The method according to claim 13, wherein the dimension of the parameter space and/or of the cluster is reduced, for example by eliminating highly correlated process parameters or by transforming vectors of the process parameters to main components within the scope of a principal component analysis (PCA), before the applying of the performance enrichment analysis.

22. The method according to claim 13, wherein visualizing the result of the first performance enrichment analysis in the form of a first dependency network, in which the process parameters from the parameter space or the clusters formed from the process parameter values, on the one hand, and the performance indicators or the external process parameter values, on the other hand, respectively form the nodes of the dependency network, and in which the connections between the nodes, evaluated with the Z-score value, represent the influence of the removed or added process parameter value on the performance indicator or the external process parameter value.

23. The method according to claim 13, wherein the performance indicator of the production process involve the following attributes: quality devaluations of the created metal product or scrap quantities; on-time schedule variances; and output quantities.

24. The method according to claim 13, wherein the process parameters from the parameter space, within and outside of the clusters, may be, for example, the following parameters: as relates to the type of product: the metal quality or the product category; as relates to the dimensions of the product: width, thickness; length, format as relates to the weight: weight per piece, coil weight, coating weight; as relates to the production capacity: mass flow, quantity; as relates to the mode of operation of the metallurgical system or one of the components thereof: pass schedule, speed.

25. The method according to claim 22, wherein additional illustrating of the results of the repeated second performance enrichment analysis in the form of a second dependency network; and combining the first and the second dependency network into one overall network with enrichment-based relationships between the clusters and the performance indicators or the external process parameter values.

26. A computer program which can be directly loaded into the internal memory of a digital computer and comprises sections of software code, with which the steps according to claim 23 are executed when the computer program is running on the computer.

Description

[0023] According to a first exemplary embodiment, it is advantageous to define a threshold value for the change between the first and the second Z-score value. This applies to the method in which the Z-score value defines the dependency of the performance indicator from the cluster and applies to the alternatively claimed method in which the Z-score value defines the dependency of an external process parameter value as a function of the cluster. The threshold value for the change of the Z-score value is preferably set such that, when the change of the Z-score value exceeds the threshold value, the influence of the cluster on the performance indicator or on the external process parameter value is considered to be relevant.

[0024] According to a further exemplary embodiment, the dimension of the parameter space is advantageously reduced respectively before the applying of the PEA. There are various methods known in the prior art for doing this. Thus, it is known to eliminate highly correlated process parameters or to transform vectors of the process parameters within the scope of a principal component analysis (PCA) to main components.

[0025] Preferably, the plurality of dependencies discovered with the aid of the PEA is transitioned into a network, the nodes of which are process parameter values and performance indicators and the connections of which represent a significant enrichment, i.e. Z-score value, preferably above the defined threshold value. This results in a visualization of the results of a performed PEA in the form of a dependency network diagram. The combining of a first network diagram, which results from applying the PEA to the original cluster, and a second network diagram, which results from applying the PEA to the changed cluster, leads to an overall network of enrichment-based relationships between process parameter values in the cluster and performance indicators or at least one external process parameter. The connections between the nodes in this overall network indicate directly and promptly implementable suggestions for troubleshooting and process optimization.

[0026] In other words, the following is claimed according to the present invention:

[0027] A function f1 evaluates a given list of orders according to the previously described method. For each identified cluster in the parameter space and for each category of the performance indicator, a Z-score value is determined which measures the predictability of the performance indicator as relates to this cluster. The result of this function is then a list of clusters with their corresponding Z-score values. The result of the first function f1 is then supplied as an input variable of a second function. In addition, a further (future) order is supplied to this second function as an input variable. First, the cluster which can be assigned to this further order is determined. Then, based on the Z-score value of this cluster, there is a determination as to whether a prediction of the performance indicator is possible for this new order. If this is the case, the prediction is output. Otherwise, it is output that there is no prediction possible for this order.